RSS Matters

You can link to the last RSS article here: SPSS' Hotfix for Windows Vista - Ed.

R Techniques: Summarizing Data By Grouping Variables

By Dr. Rich Herrington, Academic Computing and User Services, CITC

In this article we examine a few different ways of summarizing data across groups or what is referred to in as "factors" in the R language. R's model for data types consist of several different "modes" - numeric, logical, character, and factor (other data types exist, but we concern ourselves here with the type called numeric and factor. Factor variables are used in R to represent class information or "nominal" data. Converting modes, for example character modes such as "M" or "F" or "Single" or "Married", are accomplished by using the "as." conversion functions. For example, "as.factor(gender)" will convert the variable which consists of numeric 1's and 0's to a factor variable that can be used subset the data for summarization. Finally, we use various combinations of looping techniques, indexing techniques, and the "split" function to subset by groups and display the group statistics. Below we present several ways of doing this.

Several Methods for Summarizing on Factor Variables

# Create data frame with one variable that is of type "factor"
# to be used as a grouping factor

grp<-c(0,0,0,1,1,1)
grp<-as.factor(grp)

dv1<-c(10,20,30)
dv2<-c(40,50,60)

dataSet<-data.frame(grp, dv1, dv2)

attach(dataSet)

################ Version 1

# First use library "doBy", or alternatively use "split" function

# Using library "doBy"
library(doBy)
summaryBy(dv1~grp, data=dataSet, FUN=c(mean, sd), na.rm=TRUE)
summaryBy(dv2~grp, data=dataSet, FUN=c(mean, sd), na.rm=TRUE)

# Using "split" function
dataSet.list<-split(dataSet, grp)
lapply(dataSet.list, mean)
lapply(dataSet.list, sd)



################ Version  2

# Using an "apply" function only; Subsetting occurs
# within the apply calling arguments

apply(dataSet[dataSet$grp==0, c("dv1", "dv2")], 2, mean)
apply(dataSet[dataSet$grp==1, c("dv1", "dv2")], 2, sd)


################ Version 3
#
# Using "apply" With a "for loop";
# The subsetting is done by creating "index" objects that
# are then used within the apply function calling arguments

# Create "dv" index object
var.index<-c("dv1", "dv2")

# Create group index object - has the number of levels of outcome
grp.value<-c(0,1)

for (i in grp.value){
 print(i)
 print(apply(dataSet[dataSet$grp==i, var.index], 2, mean))
 print(apply(dataSet[dataSet$grp==i, var.index], 2, sd))
}


#################  Version 4

# Create a summarize function;  set parameters to send to summarize
# function;  call the summarize function in a loop and format output
# before calling summarize function

# Summarize function
my.summarize.function<-function(dataVector, grp.Vector, grp.Value){
  cat("mean\n")
  print(mean(dataVector[grp.Vector==grp.Value]))
  cat("sd\n")
  print(sd(dataVector[grp.Vector==grp.Value]))
	cat("\n\n\n")
}

# Set up the calling parameters to function "my.summarize.function"

# Select names of dv's and grouping variable

# Extract all names
var.index<-names(dataSet)

# Extract matrix with only numeric data vectors (drop grouping variable on
# column 1)
dataVectors<-dataSet[,var.index[-1]]

# Extract grouping variable (vector is column 1)
grp.Vector<-dataSet[,1]

# Declare which value of grouping variable to summarize by
grp.Value<-1


#######     Two different ways of calling my.summarize.function


# Using "apply" function only (cannot print the column names
# or the dv names when the function is called)
apply(dataVectors, 2, my.function, grp.Vector, grp.Value)


# Using a "for loop" that calls function "my.summarize.function"
# We can print the column names and dv names before calling
# "my.summarize.function";  this allows for some formatting of output

for (i in var.index[-1]){
	 cat("group = ")
   cat(paste(as.character(grp.Value)),"\n")
	 cat(paste(as.character(i)),"\n")
	 my.summarize.function(dataVector=dataVectors[, i], grp.Vector, grp.Value)
  }

Good luck and happy computing until next month. I'll leave you with the following joke that my colleagues here in the office just absolutely groaned at ..... I think they don't have a refined sense of humor.

Q: What is so hilarious about high prices with statistical software in academia?

A: Most people don't get the joke! "arrr..arrrr....arrrr...arrrr....rrrr".

Get it? It's the sound of a pirate laughing. Oh well, I'll keep working at it.

Originally published, June 2007 -- Please note that information published in Benchmarks Online is likely to degrade over time, especially links to various Websites. To make sure you have the most current information on a specific topic, it may be best to search the UNT Website - http://www.unt.edu . You can also search Benchmarks Online - http://www.unt.edu/benchmarks/archives/back.htm as well as consult the UNT Helpdesk - http://www.unt.edu/helpdesk/ Questions and comments should be directed to benchmarks@unt.edu

Return to top